aserra_phdthesis_ppt

Ph.D. Dissertation
Link Level Performance Evaluation and Link
Abstraction for LTE/LTE-Advanced Downlink
Author: Albert Serra Pagès
Thesis Advisor: Joan J. Olmos Bonafé
Department of Signal Theory and Communications
Universitat Politècnica de Catalunya
Barcelona, 5th February 2016

 Objectives
 Introduction
 Link Level Simulator for E-UTRA
 Channel Estimation Error Model (CEEM)
 E-UTRA DL Link Level Performance
 Link Abstraction for E-UTRA
 Conclusions
Outline

Objectives
 Link Level Performance Evaluation for LTE/LTE-
Advanced DL
 To develop a LTE/LTE-Advanced Link Level Simulator.
 To model the channel estimation error for link level
simulations.
 To evaluate the LTE/LTE-Advanced link level
performance for SISO-AWGN (Reference Case) and
MIMO with perfect/imperfect channel estimation.
 Link Abstraction for LTE/LTE-Advanced DL
 To propose a novel link abstraction method to predict
the BLER with good accuracy in multipath fading and
including the effects of HARQ retransmissions

Introduction
• Overview of LTE/LTE-Advanced Link Level
• Enabling technologies for LTE/LTE-Advanced

 No centralized radio
management entity
(RNC).
 All the user plane
radio functionalities
are terminated at the
eNodeB.
Overview of LTE/LTE-Advanced

Enabling technologies
OFDM MIMO
 Multiple access schemes:
OFDMA (DL) and SC-FDMA
(UL)
 Cyclic Prefix (CP) and Fast
Fourier Transform (FFT)
 Narrowband flat fading
channels (per subcarrier)
 Frequency domain equalization
 Transmit Diversity (TD), Spatial
Multiplexing (SM) and Beamforming
 Open Loop (OL), Closed-Loop (CL)
 SU-MIMO, MU-MIMO
 Up to 4 x 4 (Rel 8, 9, 10); up to 8 x 8
(Rel 11, 12)

Enabling technologies
 CQI and MCSs (QPSK, 16QAM,
64QAM)
 MIMO: PMI and RI
 HARQ with Full Incremental
Redundancy (IR)
 Max 4 Retransmissions
AMC HARQ
Link Adaptation

 Frame Structure
 Transmission Bandwidth
Frame Structure and Transmission Bandwidth
LTE Transmission Bandwidth and Resource Configuration
Channel Bandwidth 1.4 MHz 3 MHz 5 MHz 10 MHz 20 MHz
Number of RBs in the
frequency domain
6 15 25 50 100
Number of occupied
subcarriers
72 180 300 600 1200
IFFT/FFT size 128 256 512 1024 2048
Subcarrier Spacing 15 KHz / 7.5 KHz

LTE slot structure and physical resource elements
 Every 1ms (1 TTI) the resource
allocation (scheduling) and AMC
format can be changed
 Channel is almost constant for
the whole TTI.
 One Resource Block (RB)
spans 12 subcarriers in the
frequency domain.
 User data, Control channels,
Reference Signals embedded
in the lattice of Resource
Elements (REs).
 A frequency/time lattice with a
3rd dimension: spatial “layers”
(MIMO)
One radio frame = 10 ms
72subcarriers(6RBs)
(minLTEBandwidth)
One subframe =1 TTI = 1 ms (2 slots)
12subcarriers
One slot = 0.5ms
1 Resouce
Element
MIMO spatial layers

Link Level Simulator for E-UTRA
• General aspects for simulating LTE/LTE-Advanced Link
Level
• LTE/LTE-Advanced DL Link Level Simulator
• E-UTRA Transport channel processing
• E-UTRA Physical channel processing
• MIMO channel model
• MIMO Receiver Processing

Link vs. System Level Simulator
Link Level
Simulator
Transport Channel
Processing
Physical Channel
Processing
Reference BLER
System Level Simulator
Signal to Interference (SINR)
evaluation per each user and cell
H(k)
ESNR(H(k))
CQI or MCS
with
BLER(ESNR(H(k))< 10%
Results
Generate instantaneous channel (H(k)),
Path Loss Calculation and user
trajectories.
Buffering
Handover Algorithm
RRM
(Scheduling, ICIC,
ARQ, LA)
Power
Control
Traffic Generation
Average Cell Throughput
Average Cell Throughput per user
RRM evaluation statistics
Link Abstraction
EESNR, MIESM
Link to System
Mapping
BLER (ESNR) AWGN, MCS, CQI
BLERAWGN
System Level Simulator takes into account a
complete cell deployment and relies on simplified
link level look-up tables (LUTs).
Link Level Simulator simulates a single radio link with
full details between the transmitter and the receiver.
BLER, Throughput,
uncoded BER,...

Block diagram of the DL Link Level Simulator
Modular and flexible design, C/C++ off-line program.
Perfect time and frequency synchronization
IFFT/FFT & CP skipped
Simulation on the Frequency domain
Transport Channel
Processing
Physical Channel
Processing

E-UTRA Transport Channel Processing
• Turbo
Coding with
a coding
rate of 1/3
• Maximum
code block
size is 6144
bits.
• Maximum a
Posteriori
(MAP)
algorithm for
the
decoding.
Physical Channel Processing + Multipath Channel

 BICM system model as an independent and memoryless
equivalent binary channel (DMC) between a
transmitted coded bit and the received LLR.
 DMC can be properly characterized by using the Mutual
Information at bit level (MIB).
BICM Capacity
 BICM Capacity:
 Average MIB:
(bits/symbol)
where is the modulation order
(bits/LLR)

 Given a modulation scheme and a code rate r, a SNR threshold
(called BICM threshold) is the minimum SNR needed to obtain
error free transmission in AWGN conditions when that
modulation and code rate are applied assuming a capacity
approaching code.
BICM Threshold

 The DL user data is transmitted through the PDSCH in Transport
Blocks (TB).
 The transport channel PDSCH capacity in bits/subframe (Normal
Cyclic Prefix):
Transport Channel Capacity
1 ms subframe
frequency
Control region (example: 3 OFDM symbols)
Reserved for ref. signals (2 antenna port)
Parameter Description
Modulation Order in bits/symbol,
2 (QPSK), 4 (6QAM) and 6 (64QAM)
Number of allocated Resource Blocks
(RB), from 1 to 100.
Number of layers available per
codeword, from 1 to 4 for MIMO-SM
and 1 for MIMO-TD
Number of OFDM symbols used for
PDCCH, from 1 to 4.
Number of Resource Elements (RE)
reserved for pilots per RB within a
subframe

 Transport Block Size + CRC bits =>
 If code block size is higher than 6144 bits; then there is
TB fragmentation
 PDSCH payload = systematics bits + CRC bits:
 The Effective Code Rate (ECR) is the ratio of PDSCH
payload to PDSCH capacity.
Effective Code Rate (ECR)

Transport block size determination and fragmentation
 4 RBs, 1 layer/codeword, 11 PDSCH OFDM and 3 PDCCH OFDM symbols/subframe, 16 Reserved REs/RB
MCS
Index
Mod.
Mod.
Order
TBS
Index
Pilots
[RE]
PDSCH Capacity
[bits]
TB
Size
[bits]
CRC
[bits]
PDSCH
Payload [bits]
C+
[bits]
K+
[bits]
ECR
BICM
Threshold
[dB]
Threshold
delta [dB]
0 QPSK 2 0 12 960 88 24 112 1 112 0,12 -7,59
1 QPSK 2 1 12 960 144 24 168 1 168 0,18 -5,64 1,95
2 QPSK 2 2 12 960 176 24 200 1 200 0,21 -4,77 0,87
3 QPSK 2 3 12 960 208 24 232 1 232 0,24 -4,01 0,76
4 QPSK 2 4 12 960 256 24 280 1 280 0,29 -3,02 1,00
5 QPSK 2 5 12 960 328 24 352 1 352 0,37 -1,75 1,26
6 QPSK 2 6 12 960 392 24 416 1 416 0,43 -0,76 1,00
7 QPSK 2 7 12 960 472 24 496 1 496 0,52 0,37 1,13
8 QPSK 2 8 12 960 536 24 560 1 560 0,58 1,23 0,85
9 QPSK 2 9 12 960 616 24 640 1 640 0,67 2,27 1,05
10 16QAM 4 9 12 1920 616 24 640 1 640 0,33 2,33 0,06
11 16QAM 4 10 12 1920 680 24 704 1 704 0,37 2,96 0,63
12 16QAM 4 11 12 1920 776 24 800 1 800 0,42 3,85 0,89
13 16QAM 4 12 12 1920 904 24 928 1 928 0,48 4,97 1,12
14 16QAM 4 13 12 1920 1000 24 1024 1 1024 0,53 5,79 0,82
15 16QAM 4 14 12 1920 1128 24 1152 1 1152 0,60 6,86 1,07
16 16QAM 4 15 12 1920 1224 24 1248 1 1248 0,65 7,65 0,80
17 64QAM 6 15 12 2880 1224 24 1248 1 1248 0,43 8,05 0,40
18 64QAM 6 16 12 2880 1288 24 1312 1 1312 0,46 8,51 0,46
19 64QAM 6 17 12 2880 1416 24 1440 1 1440 0,50 9,42 0,91
20 64QAM 6 18 12 2880 1544 24 1568 1 1568 0,54 10,30 0,88
21 64QAM 6 19 12 2880 1736 24 1760 1 1760 0,61 11,60 1,30
22 64QAM 6 20 12 2880 1864 24 1888 1 1888 0,66 12,47 0,87
23 64QAM 6 21 12 2880 1992 24 2016 1 2016 0,70 13,35 0,88
24 64QAM 6 22 12 2880 2152 24 2176 1 2176 0,76 14,47 1,12
25 64QAM 6 23 12 2880 2280 24 2304 1 2304 0,80 15,39 0,92
26 64QAM 6 24 12 2880 2408 24 2432 1 2432 0,84 16,35 0,96
27 64QAM 6 25 12 2880 2536 24 2560 1 2560 0,89 17,40 1,05

Transport block size determination and fragmentation
MCS
Index
Mod.
Mod.
Order
TBS
Index
Pilots
[RE]
PDSCH Capacity
[bits]
TB
Size
[bits]
CRC
[bits]
PDSCH
Payload [bits]
C+
[bits]
K+
[bits]
ECR
BICM
Threshold
[dB]
Threshold
delta [dB]
0 QPSK 2 0 12 6000 680 24 704 1 704 0,12 -7,57
1 QPSK 2 1 12 6000 904 24 928 1 928 0,15 -6,25 1,32
2 QPSK 2 2 12 6000 1096 24 1120 1 1120 0,19 -5,32 0,93
3 QPSK 2 3 12 6000 1416 24 1440 1 1440 0,24 -4,05 1,27
4 QPSK 2 4 12 6000 1800 24 1824 1 1824 0,30 -2,80 1,25
5 QPSK 2 5 12 6000 2216 24 2240 1 2240 0,37 -1,65 1,15
6 QPSK 2 6 12 6000 2600 24 2624 1 2624 0,44 -0,70 0,95
7 QPSK 2 7 12 6000 3112 24 3136 1 3136 0,52 0,45 1,15
8 QPSK 2 8 12 6000 3496 24 3520 1 3520 0,59 1,27 0,82
9 QPSK 2 9 12 6000 4008 24 4032 1 4032 0,67 2,34 1,07
10 16QAM 4 9 12 12000 4008 24 4032 1 4032 0,34 2,39 0,04
11 16QAM 4 10 12 12000 4392 24 4416 1 4416 0,37 2,99 0,60
12 16QAM 4 11 12 12000 4968 24 4992 1 4992 0,42 3,84 0,85
13 16QAM 4 12 12 12000 5736 24 5760 1 5760 0,48 4,92 1,08
14 16QAM 4 13 12 12000 6456 24 6528 2 3264 0,54 5,96 1,04
15 16QAM 4 14 12 12000 7224 24 7296 2 3648 0,61 6,98 1,02
16 16QAM 4 15 12 12000 7736 24 7808 2 3904 0,65 7,66 0,68
17 64QAM 6 15 12 18000 7736 24 7808 2 3904 0,43 8,06 0,39
18 64QAM 6 16 12 18000 7992 24 8064 2 4032 0,45 8,35 0,30
19 64QAM 6 17 12 18000 9144 24 9216 2 4608 0,51 9,66 1,30
20 64QAM 6 18 12 18000 9912 24 9984 2 4992 0,55 10,50 0,84
21 64QAM 6 19 12 18000 10680 24 10752 2 5376 0,60 11,33 0,84
22 64QAM 6 20 12 18000 11448 24 11520 2 5760 0,64 12,16 0,83
23 64QAM 6 21 12 18000 12576 24 12672 3 4224 0,70 13,43 1,26
24 64QAM 6 22 12 18000 13536 24 13632 3 4544 0,76 14,50 1,08
25 64QAM 6 23 12 18000 14112 24 14208 3 4736 0,79 15,17 0,66
26 64QAM 6 24 12 18000 15264 24 15360 3 5120 0,85 16,55 1,39
27 64QAM 6 25 12 18000 15840 24 15936 3 5312 0,89 17,31 0,76
 25 RBs, 1 layer/codeword, 11 PDSCH OFDM and 3 PDCCH OFDM symbols/subframe, 16 Reserved Res/RB

E-UTRA Physical channel processing
 Lowest layer in the link level simulator
 Processes codewords according to the selected Transmission Mode (TM)
 SIC techniques blur the division between PHY channel processing and TB processing

 OFDM is combined with MIMO in order to transform the frequency-
selective nature of the MIMO wideband mobile channel model into N
parallel flat-fading subchannels (where N= number of subcarriers)
 Narrowband subcarriers allow easy equalization of multipath in
frequency domain and are also suitable to MIMO schemes.
MIMO-OFDM system model
frequency response of MIMO channel
at subcarrier

Simulating the MIMO wideband mobile channel
 A simplified stochastic matrix model based on correlation
matrices is used to generate channel coefficients:
 For each MIMO channel path:
 GWSSUS model (Gaussian
Wide-Sense Stationary
Uncorrelated Scattering)
 Doppler spectrum (Jakes low-
pass filter) and max. Doppler
frequency.
 Power Delay Profile with several
taps (3GPP models: EPA, EVA
and ETU)
MIMO correlated channel matrix =
MIMO Correlation matrix · GWSSUS channels vector
Low Medium High
α β α β α β
0 0 0.3 0.9 0.9 0.9

 Spatial Multiplexing (SM) incresases the spectral efficiency.
 Linear Detectors: ZF and MMSE. SIC techniques at the
receiver.
 Open Loop and Closed Loop Precoding.
 Global Precoded Channel Matrix:
MIMO-SM

 Open-Loop Precoding: Large Delay CDD
 Antenna port 0 is fed with x (0) (i)+x(1)(i)
 Antenna port 1 is fed with x(0)(i)-x(1)(i) for even subcarriers and with
x(1)(i)-x(0)(i) for odd subcarriers
 Closed-Loop Precoding: codebook-based
MIMO-SM Precoding

 Alamouti Space Frequency Block Coding (SFBC) in
the frequency domain.
 For two layers is pure Alamouti SFBC and the
symbols transmitted from the two antenna ports are
mapped onto each pair of adjacent subcarriers.
 MRC at the receiver (in case of more than one receive antenna)
MIMO-TD

Codeword-SIC receiver
 The link level simulator presented in this work implements a codeword-SIC
over MIMO MMSE linear receiver and takes also into account the HARQ
operation.

Channel estimation error model (CEEM)
• Objectives
• Introduction
• Reference signals in LTE downlink
• System model
• Least-squares channel estimation
• Proposed channel estimation error model
• Practical channel estimation procedure
• Validation of the channel estimation error model
• Impact of imperfect channel estimation on LTE DL
performance

 Channel estimation errors must be taken into
account to obtain realistic performance
assessments within the LTE link level simulator.
 To implement a detailed channel estimation algorithm
in the LTE link level simulator may lead to long
simulation time:
1. A Gaussian additive noise error model for channel
estimation errors is proposed and validated (CEEM).
2. Practical channel estimation methods for LTE DL
are discussed.
Objectives

 MIMO-OFDM channel estimation is
needed to obtain an accurate estimate of
the current channel matrix, per subcarrier
and symbol interval, suitable for the MIMO
processing at the receiver side.
 LTE includes pilot symbols, called
Reference Signals (RS).
 RS transmissions from the different MIMO
antennas are orthogonal, which allows
separate channel estimation for each
element of the channel matrix.
Introduction to LTE Channel Estimation
time
Freq.
1 RB
1 slot
 All pilots are QPSK
symbols following a Gold
sequence of length 31.
 The RS sequence also
carries the cell id.

System Model
 SISO-OFDM system model to study channel estimation procedures due to
the orthogonality of the DL RS. The received OFDM pilot vector is:
1
2
0 0
0 0
0
0 0 Np
C
C
C
Y H n C H n
Np → Number of pilots being processed
C → (Np x Np) Matrix with the complex pilot symbols
H → (Np x1) channel frequency response at the pilot
subcarriers
n → (Np x1) complex Gaussian noise vector with
covariance matrix 2INp

 Least-squares (LS) estimation is the baseline channel
estimation procedure. Dividing Y by the known pilots:
 LS method overestimates the average channel power
gain (G) by a factor
 Assuming that the SNR is known, we can normalize the
LS estimator to obtain the right average channel gain
Least Squares channel estimation
11 1 1ˆ ˆ
G G G
LSH H H C n
2
G
1BG 1
G
where and is the SNR of the received pilots.
22
1B= NpC C
2
=G B
1 1ˆ
LSH C Y H C n

 If we define , then and is
rewritten as:
 The estimated average channel gain is constant and can
be split into a useful contribution with variance
and a noise contribution with variance
 The parameter is always within the range [0,1] and is
a measure of how accurate the channel estimation is.
 In addition to noise there may be other sources of error
and/or improvement. In this cases we can estimate
from many realisations of and :
Channel Estimation Error Model (CEEM)
1/ 2
(1 )
12
G 1
2ˆ 1H H N
ˆH
2
1 G
2
G
2
2
ˆ
ˆ 1 1
2 GpN
H H
HˆH

Practical channel estimation procedure (I)
LS Channel Estimation
Sliding Window
Time Averaging
[Optionally]
Linear
Interpolation in
time domain
Averaging in
frequency domain
[Optionally]
Linear
Interpolation or
Wiener Filtering
in frequency
domain
time
Freq.
1 RB
1 slot
1 TTI
 Compute the LS channel
estimates at the pilot REs on
the DL frequency-time grid.
 This creates a set of LS
estimates of the channel
sampled at the pilot REs.
1 1ˆ
LSH C Y H C n

Sliding Window
Time Averaging
[Optionally]
Linear
Interpolation in
time domain
Averaging in
frequency domain
[Optionally]
Linear
Interpolation or
Wiener Filtering
in frequency
domain
time
Freq.
1 RB
1 slot
1 TTI
TTI to estimate
Practical channel estimation procedure (II)
 Optionally, perform a sliding
window time averaging of the
LS estimates at each RS
subcarrier to reduce unwanted
noise.
 The window size spans an odd
number of pilot REs, so that the
resulting average is assigned to
the RE at the centre of the
window.
 A maximum window size of 9
TTIs (17 pilots) is considered.
 A window size of 1 TTI means
no time averaging at all.

Sliding Window
Time Averaging
[Optionally]
Linear
Interpolation in
time domain
Averaging in
frequency domain
[Optionally]
Linear
Interpolation or
Wiener Filtering
in frequency
domain
time
Freq.
1 RB
1 slot
1 TTI
TTI to estimate
Practical channel estimation procedure (III)
 Perform linear interpolation in
time domain at each RS
subcarrier of the time-averaged
LS estimates to estimate the
channel for all REs of the RS
subcarriers.

Sliding Window
Time Averaging
[Optionally]
Linear Interpolation
in time domain
Averaging in
frequency domain
[Optionally]
Linear
Interpolation or
Wiener Filtering
in frequency
domain
time
Freq.
1 RB
1 slot
1 TTI
Practical channel estimation procedure (IV)
 Optionally, perform averaging in
frequency domain of the time-
averaged RS subcarriers with a
sliding window.
 The size of the window is an
odd number of RS subcarriers in
order to ensure that there is a
RS at the centre of the window.
 A window size of 1 RS
subcarrier mean no frequency
averaging at all.
TTI to estimate

Sliding Window
Time Averaging
[Optionally]
Linear
Interpolation in
time domain
Averaging in
frequency domain
[Optionally]
Linear
Interpolation or
Wiener Filtering
in frequency
domain
time
Freq.
1 RB
1 slot
1 TTI
TTI to estimate
Practical channel estimation procedure (V)
 Perform linear interpolation in the
frequency domain to estimate the
channel at each RE from the
averaged LS estimates at RS
subcarriers.
 Alternatively, instead of linear
interpolation, apply Wiener Filtering
in the frequency domain.
 is the correlation matrix of the full channel
response vector h with H in average.
 H is the complex vector that contains channel
frequency response at the pilot subcarriers.
 is the covariance matrix of the channel at
the pilot subcarriers.
H
hR hH
H
HR HH

LTE DL link level simulator parameters
Parameter Value
Carrier Frequency 2.14 GHz
Subcarrier spacing 15 KHz
Number of subcarriers
per RB
12
Number of allocated
RBs
4 RBs ( 48 subcarriers)
TTI length 1 ms
Number of OFDM
symbols per TTI
14 (11 PDSCH + 3 PDCCH)
Channel model EPA5, EVA70 and ETU300
Channel Coding Turbo code basic rate 1/3
Rate Matching and
HARQ
According to TS36.212. Max
4 IR transmissions
AMC formats (code
rate)
MCS 6 (0.44), MCS 12
(0.43), MCS 17 (0.43) and
MCS 27 (0.89)
Channel Estimation Ideal (perfect), Pilot-based,
and CEEM
Antenna scheme SISO
Acronyms if Figure Legdends
LS LS estimation
SVT Sliding window average, of size V TTIs, in time domain
LT Linear interpolation in time domain
SQF Sliding window average, of size Q pilots, in frequency
domain
LF Linear interpolation in frequency domain
WF(C) Wiener filtering in frequency domain where C is the
number of subcarrier considered for Wiener filtering
matrix
BP Pilot power boost of P dB
E-UTRA
Channel
Model
Maximum
Doppler
Frequency
Delay
Spread
(r.m.s)
50%
Coherence
Bandwidth
50%
Coherence
Time
EPA5 5 Hz 45 ns 4444 KHz 84.6 ms
EVA70 70 Hz 357 ns 560 KHz 6.0 ms
ETU300 300 Hz 991 ns 202 KHz 1.4 ms

Finding optimal parameters for EPA5 and EVA70
 Best Estimators :
 LS+S9T+LT+WF for EPA5
(averaging window of 9 TTIs)
 LS+S3T+LT+WF for EVA70
(averaging window of 3 TTIs)
 Averaging in time domain improves
substantially the estimation error
performance at low SNRs; but its
effects at high SNRs are not
significant compared to not
averaging.
 Averaging in frequency domain is
not recommended as introduces a
large error floor compared to no
averaging for medium and high
SNRs.
EPA5
EVA70

Finding optimal parameters for ETU300
 Best Estimator for ETU300:
LS+LT+WF
 Averaging in time and
frequency domain is not
recommended as introduces
a large error floor.
 Wiener Filtering Matrix of 36
subcarriers (3 RBs) WF(36)
is a good trade-off between
performance and complexity.
 For Wiener Filtering it is
assumed perfect estimation
of SNR and Power Delay
Profile.

 BLER curves are used to
validate the proposed model.
 A small gain around 1 dB can be observed
when using a pilot power boost of 6 dB for
ETU300.
Validation of CEEM
BLER (at rv=0) for EPA5 channel model and MCS 6 in
a bandwidth of 4RBs.
BLER (at rv=0) for EVA70 channel model and MCS 6
in a bandwidth of 4RBs.
BLER (at rv=0) for ETU300 channel model and MCS 6
in a bandwidth of 4RBs.

 Practical channel estimation procedures have been proposed
for different propagations conditions.
 A Channel Estimation Error Model (CEEM) has been proposed
and validated to simulate the LTE link level without the need to
process the pilot symbols or assuming ideal channel estimation.
Conclusions
 LS+S9T+LT+WF for EPA5
 LS+S3T+LT+WF for EVA70
 LS+LT+WF for ETU300

E-UTRA DL Link Level Performance
• AWGN Link Level Performance
• MIMO Performance evaluation
• E-UTRA DL Link Average Throughput

E-UTRA DL Link Level Performance
Parameter Value
per RB
12
Number of allocated
RBs
4 and 25 RBs (for AWGN)
4 (for ETU300)
TTI length 1 ms
Number of OFDM
symbols per TTI
Channel model EPA5, EVA70 and ETU300
Rate Matching and
HARQ
4 IR transmissions
AMC formats (code
rate)
According to TS 36.213
Channel Estimation Ideal (perfect)and CEEM
Antenna scheme SISO, 1x2, 2x2, 4x4
Antenna Correlation Low (LC), Medium (MC) and
High (HC)
 Study of the link
level performance:
 AWGN
 MIMO multipath
fading channel

 AWGN link level
performance is used to
determine the SNR
thresholds for link
adaptation.
 Mapping from link to
system level adopts the
form of AWGN BLER vs.
ESNR tables plus a link
abstraction method to
compute the ESNR.
AWGN Link Level Performance
 SNR (dB) needed to achieve
BLER= 10% at rv= 0 for AWGN
Channel.
 Increasing the MCS index by
one increases the SNR target
by about 1 dB.

AWGN ref. BLER curves for the LTE CQIs
 Reference BLER curves are almost regularly spaced in steps of 2dB.
 BLER slope depends on code block size.
CQI Index Modulation Code Rate
1 QPSK 0,076
2 QPSK 0,117
3 QPSK 0,189
4 QPSK 0,301
5 QPSK 0,439
6 QPSK 0,588
7 16QAM 0,369
8 16QAM 0,478
9 16QAM 0,602
10 64QAM 0,455
11 64QAM 0,554
12 64QAM 0,650
13 64QAM 0,754
14 64QAM 0,853
15 64QAM 0,926

SISO capacity and throughput in AWGN

 Given an ESNR, the code rate and the modulation order
obtained from the reported CQI, the eNodeB selects the
best MCS that maximizes the spectral efficiency.
E-UTRA Spectral Efficiency for Link Adaptation
Without HARQ the transport format
must be changed within a SINR range
of a few dB ( ~ 1 dB)
HARQ smoothes the curves, thus
allowing to use the same transform
format in a wider range of SINR

AMC Thresholds for Link Adaptation
 AMC thresholds for each MCS and each CQI in
AWGN SISO channels for 4 RBs without HARQ
2.5 dB gap between
Shannon Capacity and max
LTE Spectral Efficiency from
-5 dB to 17.5 dB of SNR

 SISO as reference
 1x2 SIMO, ZF one tap equalizer and order 2 MRC at the receiver.
 MIMO-TD: 2 x 2 or 4 x 4 MIMO with SFBC (based on Alamouti
Scheme) at the transmitter side and MRC at the receiver side.
 OL MIMO-SM: Large Delay CDD Precoding, with/without codeword-
SIC, MMSE detector.
 CL MIMO-SM: Codebook-based Precoding, with/without codeword-
SIC, MMSE detector.
Considered Transmission Modes

MIMO-SM with codeword SIC
 There is no
significant gain
with SIC in OL
with CDD
precoding.
 For CL, SIC
enhances the
non-priority
codeword link
level
performance.

E-UTRA 2x2 DL link average throughput
 2x2LC & MCS 27, Loss due
to CEEM is significantly large
for MIMO-SM schemes at high
SNRs (~4 dBs, CDD).
 2x2LC & MCS 27, MIMO-TD
at high SNRs, there is
practically no difference
between ideal and CEEM
channel estimation.
 2x2HC & MCS 27 & CEEM,
MIMO-TD outperforms MIMO-
SM; but 1x2 SIMO
outperforms MIMO-TD.
 Ideal and CEEM channel
estimation without HARQ and
MCS 15 and 27.
2x2LC
2x2HC

Link Abstraction for E-UTRA
• Introduction
• Multistate Channel
• Link Abstraction models
• Accurate Link Abstraction Method in LTE with IR HARQ
• Simulation Results

 Importance of Link Abstraction:
 Look-up table with BLER vs. channel quality thresholds:
 For CQI reporting from UE to the eNodeB.
 For fast resource scheduling at eNodeB
 Mapping from link to system level simulators.
 Link Abstraction Models:
 Exponential Effective SNR Metric (EESM)
 Mutual Information based Effective SNR Metric
(MIESM)
 To find a single scalar value, called Effective SNR
(ENSR), that summarizes the quality of the multistate
channel.
Introduction

 We need to characterize a compound multistate channel:
o Frequency selective fading: different SNR on each OFDM subcarrier.
o LLR combination of different HARQ retransmissions.
o Unequal error protection in 16QAM and 64QAM.
 Our goals are:
 To propose an HARQ aware link abstraction model for LTE/LTE-
Advanced to predict the BLER with good accuracy in multipath fading
and including the effects of HARQ retransmissions based on Mutual
 To assess the BLER prediction accuracy for SISO and 2x2 MIMO DL.
Introduction

Link Abstraction Models: EESM and MIESM
 ESNR ( ): a single scalar value
that summarizes the multistate
channel quality:
 MIESM model uses the sigmoid
function I(·) = Average Mutual
 EESM model uses I( ) = 1-exp( /β),
which is easier to compute (closed
expression).
 α1 and α2 need to be adjusted using
a link level simulator. A possible
simplification is to take α1 = α2 = β
11 2
1 N
eff k
k
I I
N
MIB is modulation specific
and has not a closed
expression. MIB for LTE
modulation schemes:

 MIB computed numerically using:
 For BPSK:
Mutual Information at Bit level (MIB)
0.12
( ) ( ) 1 exp( /10 )BPSKMIB I
|
, 2 2 |
0,1 |
|
( | ) 1 2
( , ) log log ( | )
( | )( ) 2
1
( | )
z b
b z z b i
i z b iz
z b i
f z b
MI b z E f z b i dz
f z b if z
f z b i
Channel z=LLR(b)b
b {0,1}

 In a multistate channel the average received bit information rate is:
Effective SNR
*
1
1 bitsN
i
ibits
r
N
AWGNBLER( ) BLER ( , )ESNR ESNR r
MI carried by bit i

 We start considering only BPSK modulation:
 Within U0 some bits may be repeated up to 3 times (for rate << 1/3):
 We need to approximate the exact values by their mean values:
 Finally, for BPSK with frequency selective fading and bit repetitions:
Average bit information rate for first transmission
00
1
* ( )BPSK i
i U
r I
00 0
* 31 2
0 0 0
( ) ( ) ( )k k l k l mBPSK
r I I I
0 0 0
1 2 3
*
,1 ,1 ,2 ,1 ,2 ,3
0
1
( ) ( ) ( )i i i i i iBPSK
i U i U i U
r I I I
0
1
,10
11
1 1
( ) ( ) ( )
SCN
i k k
ki U SC
I I I
N
0
2
,1 ,20 2
1 12
1 1
( ) ( ) ( )
SC SCN N
i i k l k l
k li U SC
I I I
N
0
3
,1 ,2 ,30 3
1 1 13
1 1
( ) ( ) ( )
SC SC SCN N N
i i i k l m k l m
k l mi U SC
I I I
N

 Those terms are too complex to compute for high system bandwidth:
 We use a property of function I(·):
 For example:
 Interleaving creates an independent fading on every transmission, so we
can multiply the averages.
 Finally, only these terms need to be computed:
Simplified computation of r*
2
1 1
1
( ) ( )
SC SCN N
k l k l
k lSC
I I
N
3
1 1 1
1
( ) ( )
SC SC SCN N N
k l m k l m
k l mSC
I I
N
1 2 1 1 2 1, , , 1 , , ,n n n n n nI x x x I x I x I x x x
2 , ( ) 1 ( ) ( )k l k l l l kI I I I I
0.12
( ) ( ) 1 exp( /10 )BPSKMIB I

Capturing the effects of unequal error protection
 Proposed procedure:
1. Obtain the BPSK
equivalent SNRs of the two
bit channels:
2. Add the BPSK equivalent
SNRs and obtain the global
MIB:
 The model is valid if the two
transmissions happen in the
same or in different redundancy
versions.
Example with 16QAM

 We need to approximate the exact values by their mean values:
 Simplified computation of averages:
 Within U0 some bits may be repeated up to 3 times (for rate << 1/3):
Average bit information rate for first transmission
2( ) ( ) ( ), ( ) ( ( )) 1 ( ( )) ( ( ))kA k lA l kA k lA l lA l lA l kA kI I I I I

 The set of bits that have been received (at least once) after 2nd round (U1)
is decomposed into 15 subsets:
 Computation of r* is a direct extension of the expression for 1st round:
Extension to 2nd H-ARQ round

 Reference BLER curves are obtained by simulating with
the mother code rate and the 3 possible modulation
schemes:
A reduced set of AWGN reference BLER curves

A reduced set of AWGN reference BLER curves
 The “HARQ Effective code rate (reff)“ is defined from the point of
view of the decoder:
 For the first round reff = r (rate of the MCS). reff decreases with every
HARQ round.
 For reff > 1/3 there is a coding gain
reduction with respect to the reference
BLER:
 The reference BLER is shifted dB
1/3effr
1 1
[dB] ( ) (1/3) ( 0)effMIB r MIB
 For reff = 1/3: There is no coding gain with respect to the reference
BLER. There is only an energy gain, which is captured by how is
computed the ESNR.
 The reference BLER curves (code rate=1/3) are OK.

E-UTRA DL Link Level Parameters
Parameter Value
per RB
12
Number of allocated
RBs
1 and 25 RBs (for ETU300)
TTI length 1 ms
Number of OFDM
symbols per TTI
Channel model ETU300
Rate Matching and
HARQ
4 IR transmissions
AMC formats (code
rate)
According to TS 36.213
Channel Estimation Ideal (perfect)
Antenna scheme SISO, 2x2 MIMO
Antenna Correlation Low (LC) and High (HC)
 MIMO multipath fading
channel
 Channel is perfectly
known at each
subcarrier
 ETU300 channel
model.
 LC and HC antenna
correlation
 Transmissions Modes:
 SISO
 2x2 MIMO-SM with
CDD precoding
 2x2 MIMO-TD

BLER prediction results: SISO
1st. HARQ round 2nd. HARQ round
3rd. HARQ round 4th. HARQ round

BLER prediction results: MIMO 2x2LC

BLER prediction results: MIMO 2x2 HC

Conclusions of the proposed Link Abstraction Method
 Observing the simulation results, there is a good
match between the predicted and simulated BLER
for ETU300 channel model, ideal channel estimation,
5MHz bandwidth and SISO, 2x2 MIMO-SM with CDD
precoding and 2x2 MIMO-TD.
 Advantages of the proposed method:
 No calibration.
 Only three reference BLER curves.
 All the effects of the multistate channel in highly
selective fading are captured.

Conclusions and Open Issues
• Conclusions
• Achieved goals
• Open Issues
• Dissertation Publications

 Channel Estimation:
 CEEM is validated on link level simulations.
 The Wiener filter in the frequency domain leads to
low channel estimation error.
Conclusions
 DL Link Level Performance:
 SNR thresholds for AMC link adaptation in AWGN
conditions, spaced aprox. 1 dB.
 Codeword-SIC useful for the non-priority codeword of
CL MIMO-SM
 It has been proposed a novel link abstraction method
that can predict the BLER with good accuracy in
multipath fading and including the effects of HARQ
retransmissions.

1. Development of a E-UTRA DL link level
simulator.
2. Proposal and validation of Channel
Estimation Procedures and a Channel
Estimation Error Model (CEEM).
3. Evaluation of the E-UTRA DL link level
performance for SISO and MIMO TMs with
perfect and imperfect channel estimation.
4. Proposal and validation of a novel link
abstraction method for E-UTRA including the
effects of IR HARQ retransmissions.
Achieved goals

 To extended the link level simulator to include: UL, the full
set of Transmission Modes and optimization of the CL
precoding selection without assuming perfect selection per
subcarrier.
 To use a more complex wideband MIMO channel model
which models the geometry of the scattering in a stochastic
way.
 To extended the CEEM LUTs to other type of reference
signals.
 To extended the AWGN link level performance to the new
256QAM coding schemes (LTE Release 12)
 To extended the Link Abstraction method to other DL
transmission modes, such as MU-MIMO and UL
transmission modes.
Open Issues

Dissertation Publications
 For the development of a LTE link level simulator:
 Joan Olmos, Albert Serra, Silvia Ruiz, Mario García-lozano, and David
Gonzalez. Link Level Simulator for LTE Downlink. In COST 2100 TD(09)779,
2009.
 For the study of optimum link abstraction methods:
Gonzalez. Exponential Effective SIR Link Performance Model for LTE
Downlink. COST 2100 TD09)874, 2009,
Gonzalez. Exponential Effective SIR Metric for LTE Downlink. 20th IEEE
International Symposium On Personal, Indoor and Mobile Radio
Communications, pages 900904, 2009.
 Joan Olmos, Albert Serra, Mario García-Lozano, Silvia Ruiz, and David Pérez
Díaz De Cerio. Simulation of LTE IR H-ARQ at System Level Using MIESM
Error Prediction. IC1004 TD(11)02072, 2011.
 Joan Olmos, Albert Serra, Silvia Ruiz, and Imran Latif. On the Use of Mutual
Information at Bit Level for Accurate Link Abstraction in LTE with
Incremental Redundancy H-ARQ. In IC1004 TD(12)05046, 2012.

Dissertation Publications
 For the study of how to model channel estimation error for link
level simulations:
 Albert Serra, Joan Olmos, and Maria Lema. Modelling Channel Estimation
Error in LTE Link Level Simulations. IC1004 TD(12)03067, 2012.
 For the definition of reference scenarios for LTE/LTE-
Advanced link level simulations:
 Joan Olmos, Albert Serra, and Silvia Ruiz. On the Definition of Reference
Scenarios for LTE-A Link Level Simulations within COST IC1004. In IC1004
TD(13)06043, 2013.
 Additionally, the work of this dissertation has contributed to the
simulation and study of the LTE system level:
 David González, Silvia Ruiz, Joan Olmos, and Albert Serra. System Level
Evaluation of LTE Networks with Semidistributed Intercell Interference
Coordination. In IEEE 20th International Symposium on Personal, Indoor and
Mobile Radio Communications, 2009.
 David Gonzalez, Joan Olmos, Silvia Ruiz, and Albert Serra. Downlink Inter-Cell
Interference Coordination and Scheduling for LTE Featuring HARQ over
Multipath Fading Channel. pages 15, 2009.
 David Gonzalez, Silvia Ruiz, Joan Olmos, and Albert Serra. Link and System
Level Simulation of Downlink LTE. In COST 2100 TD(09)734, 2009.

albert.serra@telecos.cat
Questions ?

albert.serra@telecos.cat
Moltes Gràcies

aserra_phdthesis_ppt

Empfohlen

Empfohlen

Weitere ähnliche Inhalte

Andere mochten auch

Andere mochten auch (13)

Ähnlich wie aserra_phdthesis_ppt

Ähnlich wie aserra_phdthesis_ppt (20)

aserra_phdthesis_ppt